Moving Object Detection Based On Computer Vision

Intelligent monitoring system has been widely used in all aspects of people's production and life,from home to public places.It can not only help people protect their own rights and interests,maintain property safety,but also help maintain the security and stability of the whole society.Intelligent monitoring system involves moving object detection,classification,tracking and behavior understanding.This thesis focuses on the research of moving object detection in intelligent video analysis,in order to achieve the purpose of accurate detection of moving objects in complex video scene,and better ensure the correctness of intelligent monitoring system decision.In traditional Vi Be algorithm,there is ghost phenomenon of moving object or static object moving suddenly.In this thesis,three frame difference method,improved Canny operator and improved adaptive Vi Be algorithm are combined to detect moving object.Firstly,the traditional Canny edge detection operator is improved,that is,the Gaussian filter is changed to bilateral filter;The two direction template of Sobel operator is added to eight direction template;The fixed double threshold method is changed to the maximum inter class difference method.Compared with the traditional Canny algorithm,the number and accuracy of the target edges detected by the improved Canny algorithm are improved by 5%and 2% respectively.Secondly,on the basis of the traditional Vi Be algorithm,adaptive threshold and adaptive update rate are used to achieve the detection of moving objects,which improves the shortcomings of the traditional Vi Be algorithm,such as the slow speed of ghost elimination and the increase of false detection rate caused by the fixed threshold and background update rate.The experimental results show that,compared with the traditional Vi Be algorithm,the combination of the three frame difference method,the improved Canny operator and the improved adaptive Vi Be algorithm improves the comprehensive accuracy and recall rate by 3.0%.In the traditional method of moving shadow detection based on HSV color space,the moving shadow of the object will be mistakenly detected as the object in the illumination.In this thesis,color features and texture features are fused to detect the moving shadow.Experimental results show that compared with the moving shadow detection based on HSV color space,the proposed method improves the detection rate and resolution by 8%.To sum up,the work of this thesis is to improve the comprehensive accuracy and recall rate of moving object detection compared with the traditional Vi Be algorithm;Compared with the HSV color space method,the comprehensive shadow detection rate and resolution of moving shadows are improved.